A try for emebdding model:
The method is the same as the stella-v2, I just extend the length of the context on tao.(I found if you want to use the fully-8k context, you maybe need to convert the model to float32).
Now I'm working on the tao-v2, It will have a different sturcture.
I will release tao-v2 as fast as I can.
Thank you to the open source community.
- Downloads last month
- 10
Evaluation results
- cos_sim_pearson on MTEB AFQMCvalidation set self-reported47.336
- cos_sim_spearman on MTEB AFQMCvalidation set self-reported49.940
- euclidean_pearson on MTEB AFQMCvalidation set self-reported48.120
- euclidean_spearman on MTEB AFQMCvalidation set self-reported49.940
- manhattan_pearson on MTEB AFQMCvalidation set self-reported48.074
- manhattan_spearman on MTEB AFQMCvalidation set self-reported49.892
- cos_sim_pearson on MTEB ATECtest set self-reported50.976
- cos_sim_spearman on MTEB ATECtest set self-reported53.112
- euclidean_pearson on MTEB ATECtest set self-reported55.120
- euclidean_spearman on MTEB ATECtest set self-reported53.112
- manhattan_pearson on MTEB ATECtest set self-reported55.097
- manhattan_spearman on MTEB ATECtest set self-reported53.107
- accuracy on MTEB AmazonReviewsClassification (zh)test set self-reported40.804
- f1 on MTEB AmazonReviewsClassification (zh)test set self-reported39.011
- cos_sim_pearson on MTEB BQtest set self-reported62.844
- cos_sim_spearman on MTEB BQtest set self-reported65.541
- euclidean_pearson on MTEB BQtest set self-reported64.088
- euclidean_spearman on MTEB BQtest set self-reported65.541
- manhattan_pearson on MTEB BQtest set self-reported64.094
- manhattan_spearman on MTEB BQtest set self-reported65.554